Classification Of Land Use And Land Cover In The Alpine And Gorge Region Based On RS

In this paper, the ASTER remote sensing data classification result, which extractd by the methods of unsupervised, supervised and integrated threshold was analyzed by the support of 3S technology and fieldwork in the alpine and gorge region in the Ganzi state. Results showed that:(1)Has taken format transforms, geometry correction, image merge and image enhancement to fulfill the data processing with the software of ERDAS. Through the compare to the result,it has finally choosed the preferable mage as the base data for further analysis. In addition, the result of eye-estimation was verified by using qualitative and quantitative method to determine the surface features. In quantitative estimation, the promiscuous type was estimated by the method of Back Propagation Neural Network (BPNN).The result showed that BP network was convergent in the research, and the error precision could attained the initialization. Except of cultivated land, the correct ratio of classification was relative exactly. Each of them was forest (95.83%), rock (93.33%), snow(90.00%),grass(79.17%),residentialarea(87.50%),river(84.00%),cultivated land(25 %) in respectively. Integrated threshold classification approach was supported by GIS, which was based on the result of classification experiment and actual situation about land use and land cover in this area. By the support of knowledge formula, the information was extracted in research.(2)This paper compared the result of different classify method, through combining point and area. Among the methods, the precision of unsupervised, supervised and integrated threshold was 36.56%, 60.85% and 71.70% respectively. In the result of unsupervised classification, the extracting effect of forest was best, cultivated land and residential area were worse. Supervised classification, it was easy to identifying snow and river, through the texture and chroma characteristics. Integrated threshold classification, snow was most well-marked type in the research area. The drawing precision and user precision was 90.91% and 100% respectively. The classification precision was not only concerned with classification rule and parameter, but also training samples.Error area statistics, the expression trend was similar to point. Among the methods, the total precision was unsupervised classification lower than supervised classification and supervised classification lower than integrated threshold classification. In unsupervised classification, the type of forest and snow was most well-marked, and the absolute area was 78.50 km~2 and -3.12 km~2, relative error was 12.26% and 2.04% respectively. Cultivated land and residential area was most hard to identified, the absolute area was 153.71 km~2 and 47.28 km~2 , and corresponding relative error was 6445.15% and 3528.00%. In the result of integrated threshold classification, forest and other type were most well-marked, the the absolute area was 7.30 km~2 and -147.79 km~2, the relative error was 1.15% and -71.45%.(3)In a word, cultivated land was the most difficult to identify, and river and snow were most well-marked types in this research. Dduring the transformation form qualitative to quantitative method, the classification precision and reliability was improved. Compared to the method of unsupervised and supervised classifition, the approach of integrated threshold had obvious advantage, and more adapt to be used in the alpine and gorge region.